AI in Medical TourismAttracting International Patients on My Own with AI — How Far Can AI Go?
Key points
- Genie is ChatGPT, and Theo is Claude Code. In our work attracting international patients for MyClinic, Genie mainly produces content, and Theo codes the service.
- From the raw footage, Genie picked out segments and made a main video of about 3 minutes 35 seconds, plus five short-form videos re-edited from it.
- Even when Genie edits a video well, it still needs a human hand before it can actually be used.
- There is no longer any doubt that AI agents can take over people's work and produce high-quality results in every part of a business. Now I want to move to the stage of understanding how much time working with AI has saved me, and how much the cost of attracting international patients has changed, along with bookings and revenue.
I work with Genie and Theo. Genie is ChatGPT, and Theo is Claude Code. In our work attracting international patients for MyClinic, Genie mainly produces content, and Theo codes the service.
Genie and Theo do a lot of work. They edit interview videos and upload them to YouTube, and they make short-form videos and carousel posts and post them on Instagram. They reply to messages directly, not only in English but also in Chinese, Thai, and Japanese. Now that we have Genie and Theo, and ChatGPT’s Dots has come out as well, they share everyday office work with me, too.
One AI agent was doing the work of 4–5 employees.
From the raw footage, Genie picked out segments and made a main video of about 3 minutes 35 seconds, plus five short-form videos re-edited from it. It divided the content into parts such as pre-visit guidance, experiences with the medical staff, interpretation, and the stay and getting around, and added English and Chinese subtitles. It also cleaned up screens where our old brand still showed, and prepared the covers and the post copy.
After content production, they also answer customer inquiries directly. Theo routes inquiries from Instagram and Facebook Messenger to the MyClinic site and sends guidance to real customers based on our dataset. When we guide customers on preparing for their visit and on the clinic’s location, I review the Korean content first. Then Genie reads the earlier conversation and the latest messages, checks that the same guidance hasn’t already been sent, and sends it. It also shows me the English sentences it sent, along with their Korean meanings.
But it didn’t come together with a single “click.”
Even when Genie edits a video well, it still needs a human hand before it can actually be used. The first time I worked on videos with Genie, I looked at the finished video and cover and had Genie fix the title, the line breaks, and the placement of the people. We had to bring out what the customer was told and how they felt, while making sure we didn’t add anything that wasn’t in the video as fiction.
We had to readjust the aspect ratio and the cover for how they appear on Instagram, and we had to check both how things looked in the edit file and how they actually looked on the channel. Genie kept learning too, uploading the revised files and checking that they were publicly visible and playable.
The written posts needed even more fixing than the videos. The drafts read as if a third party were describing what was in the video. I had Genie reread the full transcript, and I passed along details I had checked separately. After that step, even with words from the same interview, the wording changed depending on who had experienced and explained what.
In the videos, the titles had to match what was actually said. In the posts, we had to restore missing parts of the story and correct who was identified as speaking. While Genie did the editing and translation, I reviewed the results and flagged what still needed checking and fixing. My job didn’t end with handing over the source material and hitting “click.”
Even with the authority, it stopped
Once, Genie held off even though it could have run ads within a budget I had already approved. The reason was that no one had confirmed whether the feature that automatically alerts the person in charge about ad inquiries was working properly. It was Genie’s judgment, meant to make sure no inquiry was missed. But in effect, the work stopped until the automatic alert for the person in charge was ready. After that, whenever Genie said it was waiting, I first had to check what it was waiting for.
Adding more materials and checks made the work complicated
When I first handed content production to Genie, I had to let it read our existing marketing materials. It needed to understand the direction of the company’s service and our customers to create content properly. And the content also had to come out consistently, in a set format. So I kept the materials in my Obsidian vault, where I keep my personal and work records, and also in Google Drive. I wrote up standards and procedures as documents so I could check the work, and when it still wasn’t clear what needed to be done, I started a log in Notion. Every time a problem came up, we added one more place to record things and one more procedure to check.
Handing off work was complicated too. At one point, we even made this the default: Genie made a request to Theo, Theo confirmed, and Genie checked again. Even tasks Genie could finish with its existing tools picked up a handoff step between the two AIs. Slack kept pinging, and my own work kept stalling. In the end, I apologized to the two AIs: “I think that in working with Genie and Theo, I’ve actually made the way we operate more confusing.”
In the end, starting over by sorting things out
So I decided to start by laying out the work it takes to attract international patients. To answer customers who come in after seeing our content, we need to understand their questions, check what’s needed with the clinic, and guide them so they can prepare for their visit. Within this process, I separated which tasks Genie can finish and where Theo’s development work is needed. We also decided how far each task must go before the next person in charge can pick it up.
I also sorted out where we keep records. The company’s standards and approved deliverables go in Google Drive, and the detailed work history and current to-dos go in the Obsidian vault. I set it up so the materials needed for the actual work and the current status are easy to find.
Start with the customer and the goal
There was a similar gap in content planning. Posts and videos were being made, but there wasn’t enough thinking about who the content was for, what message it would deliver, and what action it should lead to.
So I updated our standards: before production, we research what customers are asking and what others are covering, and connect it to our goals. After deciding what to cover, we choose whether it will be a post or a video, and which channel to publish it on. After it goes live, we need to check how far people progressed toward the intended action and use that in the next plan. Handing research and planning to Genie means reviewing the reasoning behind these choices together as well.
To attract international patients with AI.
There is no longer any doubt that AI agents can take over people’s work and produce high-quality results in every part of a business. Now I want to move to the stage of understanding how much time working with AI has saved me, and how much the cost of attracting international patients has changed, along with bookings and revenue. I also want to follow a single inquiry that came in through our content, from confirming details with the clinic to the customer’s decision to visit. I want to find out what information was needed, who prepared the answers, and where work stopped because something had to be checked again. As those records build up, maybe I’ll learn more about how far AI can be used in attracting international patients.
